Research And Implementation Of Elasticity Supporting Mechanism For Distributed Applications

As a network-based computing method, cloud computing has been receiving at-tention from academia and industry. Platform of cloud computing form a virtual re-sourcepoolbyconnectingenormousamountofcomputingandstorageresourcesthroughnetwork. Applications that are deployed on such platform cope with fuctuation ofworkload by demanding resources provided by the platform. The ability to utilize re-sources on demand is called elasticity. Diferent from scalability, elasticity requiresnot only increase in utilization of additional resources to handle extra workload, butalso the release of assigned resources when workload reduces. Supporting elastic ap-plications is the major character of cloud computing in comparison to other distributedcomputing method such as grid computing.Academic and industry have conducted series of researches from the perspectiveof platform. Cloud computing is a form of distributed computing, hence it inheritsthe two approaches to acquire additional resources: scaling up and scaling out. Toscale up is to assign more power to specifc nodes, and to scale out is to expand byutilizing more nodes. Cloud platform support elasticity accordingly. PaaS(Platform asa Service) platforms mask the distributed infrastructure from applications, and providescaling up option. IaaS(Infrastructure as a Service) platforms provide the interfaces tocreation and deletion of virtual machines, and provide the option of scaling out. Thispaper focuses on the latter case.Without proper architecture, an application deployed on IaaS platform will notachieve elasticity even with abundant resources. This paper conducts research fromthe perspective of application, and proposes the following methodology: frstly clas-sify applications by their business logic, secondly propose a pattern for this class of applications, thirdly design and implement the architecture and elasticity supportinglibrary of the supporting platform, lastly verify the pattern through experiments.The above methodology is applied during the case study of Web1.0and Web2.0applications. The fact that contents of Web1.0are purely generated on server side in-dicates that static webpage is the nature solution. Publisher-Distributor-Author Patternis proposed to separate the generation and publicity of contents, and to deal with thefuctuation of workload. Experiments are conducted to verify this pattern.User Generated Content is the essence of Web2.0. Processor-State-Persistent Pat-tern is proposed to separate the process of business logic computing, session data man-aging and persistent data managing. Modifcation of Consistent Hashing is brought upduring the implementation of Web2.0platform and elasticity supporting library. Theresults of the above two case studies proves the validity of methodology, as well as theefectiveness of the two proposed patterns.